feat: Add Claude Sonnet 4.6 Thinking model support#496
feat: Add Claude Sonnet 4.6 Thinking model support#496sk0x0y wants to merge 1 commit intoNoeFabris:mainfrom
Conversation
WalkthroughThis pull request adds support for the Claude Sonnet 4.6 Thinking model (
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Greptile SummaryAdded support for Claude Sonnet 4.6 Thinking model with extended thinking capabilities. The implementation follows the established pattern from Key changes:
Issues found:
Confidence Score: 3/5
Important Files Changed
Flowchart%%{init: {'theme': 'neutral'}}%%
flowchart TD
A[User Request: antigravity-claude-sonnet-4-6-thinking] --> B{Has Variant Suffix?}
B -->|No suffix| C[resolveModelWithTier]
B -->|With -low/-medium/-high| C
C --> D[Strip 'antigravity-' prefix]
D --> E[Extract tier from suffix]
E --> F{Tier extracted?}
F -->|Yes: low/medium/high| G[Map to thinking budget]
F -->|No tier| H[Use default max budget: 32768]
G --> I[low: 8192<br/>medium: 16384<br/>high: 32768]
H --> J[claude-sonnet-4-6-thinking]
I --> J
J --> K[Return ResolvedModel with thinkingBudget]
style A fill:#e1f5ff
style J fill:#ffe1e1
style K fill:#e1ffe1
Last reviewed commit: 55dd862 |
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
| | `antigravity-gemini-3.1-pro` | low, high | Gemini 3.1 Pro with thinking (rollout-dependent) | | ||
| | `antigravity-gemini-3-flash` | minimal, low, medium, high | Gemini 3 Flash with thinking | | ||
| | `antigravity-claude-sonnet-4-6` | — | Claude Sonnet 4.6 | | ||
| | `antigravity-claude-sonnet-4-6-thinking` | low, medium, high | Claude Sonnet 4.6 with extended thinking | |
There was a problem hiding this comment.
Variants list is inconsistent with the actual model definition. The model defines low and max variants (matching antigravity-claude-opus-4-6-thinking on line 125), not low, medium, high.
| | `antigravity-claude-sonnet-4-6-thinking` | low, medium, high | Claude Sonnet 4.6 with extended thinking | | |
| | `antigravity-claude-sonnet-4-6-thinking` | low, max | Claude Sonnet 4.6 with extended thinking | |
Prompt To Fix With AI
This is a comment left during a code review.
Path: README.md
Line: 124
Comment:
Variants list is inconsistent with the actual model definition. The model defines `low` and `max` variants (matching `antigravity-claude-opus-4-6-thinking` on line 125), not `low, medium, high`.
```suggestion
| `antigravity-claude-sonnet-4-6-thinking` | low, max | Claude Sonnet 4.6 with extended thinking |
```
How can I resolve this? If you propose a fix, please make it concise.
Additional Comments (1)
Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time! Prompt To Fix With AIThis is a comment left during a code review.
Path: src/plugin/config/models.test.ts
Line: 52-57
Comment:
Consider adding test case for `antigravity-claude-sonnet-4-6-thinking` variants to match the pattern of the Opus test.
<sub>Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!</sub>
How can I resolve this? If you propose a fix, please make it concise. |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
🧹 Nitpick comments (4)
src/plugin/transform/gemini.test.ts (1)
615-618: LGTM — the new assertion is correct;isImageGenerationModelwon't match the "thinking" suffix.Optional nit: the test title says "thinking variants" (plural) but only tests one. If you want exhaustive coverage you could also add
"claude-sonnet-4-6-thinking-low", but this is not necessary given the purely substring-based implementation.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/plugin/transform/gemini.test.ts` around lines 615 - 618, Test title says "thinking variants" (plural) but only asserts one variant; update the spec in src/plugin/transform/gemini.test.ts for the isImageGenerationModel checks: either change the it(...) title to singular ("thinking variant") or add an additional assertion such as expect(isImageGenerationModel("claude-sonnet-4-6-thinking-low")).toBe(false) so the title and assertions match; reference the isImageGenerationModel calls in the failing test block and adjust accordingly.docs/MODEL-VARIANTS.md (1)
163-172: Optional: add Sonnet 4.6 Thinking tier-suffixed names to the "Tier-Suffixed Names" inventory.The resolver supports
antigravity-claude-sonnet-4-6-thinking-{low,medium,high}(same path as the Opus variants), but this section only lists Opus entries. Users who rely on tier-suffixed model strings won't know these are accepted.📝 Suggested addition
- `antigravity-claude-opus-4-6-thinking-low` - `antigravity-claude-opus-4-6-thinking-medium` - `antigravity-claude-opus-4-6-thinking-high` +- `antigravity-claude-sonnet-4-6-thinking-low` +- `antigravity-claude-sonnet-4-6-thinking-medium` +- `antigravity-claude-sonnet-4-6-thinking-high`🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@docs/MODEL-VARIANTS.md` around lines 163 - 172, Update the "Tier-suffixed model names" list to include the Sonnet 4.6 Thinking variants that the resolver supports: add entries for antigravity-claude-sonnet-4-6-thinking-low, antigravity-claude-sonnet-4-6-thinking-medium, and antigravity-claude-sonnet-4-6-thinking-high (matching the same path as the Opus variants) so users relying on tier-suffixed strings see they are accepted.src/plugin/config/models.test.ts (1)
52-57: Consider extending the variant assertion to cover the new Sonnet 4.6 Thinking model.The "defines thinking budget variants for Claude thinking models" suite only asserts against
antigravity-claude-opus-4-6-thinking. Since the PR also addsantigravity-claude-sonnet-4-6-thinkingwith the same variant shape, this is a natural fit to validate here. Without it, the variant definition for the new model is exercised only indirectly through the resolver tests.🛠️ Suggested addition
it("defines thinking budget variants for Claude thinking models", () => { expect(getModel("antigravity-claude-opus-4-6-thinking").variants).toEqual({ low: { thinkingConfig: { thinkingBudget: 8192 } }, max: { thinkingConfig: { thinkingBudget: 32768 } }, }); + + expect(getModel("antigravity-claude-sonnet-4-6-thinking").variants).toEqual({ + low: { thinkingConfig: { thinkingBudget: 8192 } }, + max: { thinkingConfig: { thinkingBudget: 32768 } }, + }); });🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/plugin/config/models.test.ts` around lines 52 - 57, Add an assertion for the new Sonnet thinking model so its variants are directly validated: update the "defines thinking budget variants for Claude thinking models" test (which calls getModel) to also expect getModel("antigravity-claude-sonnet-4-6-thinking").variants toEqual the same shape as the Opus one ({ low: { thinkingConfig: { thinkingBudget: 8192 } }, max: { thinkingConfig: { thinkingBudget: 32768 } } }); this ensures both antigravity-claude-opus-4-6-thinking and antigravity-claude-sonnet-4-6-thinking are explicitly checked.src/plugin/transform/model-resolver.test.ts (1)
121-137: Add test coverage for Sonnet 4.6 Thinking alias variants and tier-suffixed backward compatibility.The new
gemini-claude-sonnet-4-6-thinking-low/medium/high → claude-sonnet-4-6-thinkingaliases in model-resolver.ts (lines 55–57) are untested. Additionally, the backward-compatibility suite coversclaude-opus-4-6-thinking-low/medium/highbut lacks the equivalent Sonnet tests.Add tests mirroring the existing Opus coverage:
🛠️ Suggested additions
In the
"Claude Sonnet 4.6 (non-thinking)"describe block (around line 157–162):+ it("gemini-claude-sonnet-4-6-thinking-low alias resolves to claude-sonnet-4-6-thinking with low budget", () => { + const result = resolveModelWithTier("gemini-claude-sonnet-4-6-thinking-low"); + expect(result.actualModel).toBe("claude-sonnet-4-6-thinking"); + expect(result.thinkingBudget).toBe(8192); + });In the
"backward compatibility"describe block (after line 263):+ it("claude-sonnet-4-6-thinking tier-suffixed models work without variant config", () => { + const lowResult = resolveModelWithVariant("claude-sonnet-4-6-thinking-low"); + expect(lowResult.thinkingBudget).toBe(8192); + + const highResult = resolveModelWithVariant("claude-sonnet-4-6-thinking-high"); + expect(highResult.thinkingBudget).toBe(32768); + });🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/plugin/transform/model-resolver.test.ts` around lines 121 - 137, Add test cases to cover the new Sonnet 4.6 thinking aliases: ensure resolveModelWithTier("antigravity-claude-sonnet-4-6-thinking") and the tier-suffixed aliases ("gemini-claude-sonnet-4-6-thinking-low/medium/high") map to actualModel "claude-sonnet-4-6-thinking", have thinkingBudget 32768, isThinkingModel true, and quotaPreference "antigravity". Mirror the existing Opus tests: add the non-thinking Sonnet test into the "Claude Sonnet 4.6 (non-thinking)" describe block and add the tier-suffixed backward-compatibility tests into the "backward compatibility" describe block so resolveModelWithTier and the new aliases in model-resolver.ts are exercised.
📜 Review details
Configuration used: Repository UI
Review profile: CHILL
Plan: Pro
Cache: Disabled due to data retention organization setting
Knowledge base: Disabled due to data retention organization setting
📒 Files selected for processing (10)
README.mddocs/ANTIGRAVITY_API_SPEC.mddocs/MODEL-VARIANTS.mdsrc/plugin/config/models.test.tssrc/plugin/config/models.tssrc/plugin/transform/claude.test.tssrc/plugin/transform/cross-model-sanitizer.test.tssrc/plugin/transform/gemini.test.tssrc/plugin/transform/model-resolver.test.tssrc/plugin/transform/model-resolver.ts
🧰 Additional context used
🧬 Code graph analysis (4)
src/plugin/transform/cross-model-sanitizer.test.ts (3)
src/plugin/transform/model-resolver.ts (1)
getModelFamily(275-284)src/plugin/transform/index.ts (2)
getModelFamily(25-25)getModelFamily(65-65)src/plugin/transform/cross-model-sanitizer.ts (1)
getModelFamily(30-34)
src/plugin/transform/claude.test.ts (2)
src/plugin/transform/index.ts (1)
isClaudeThinkingModel(35-35)src/plugin/transform/claude.ts (1)
isClaudeThinkingModel(34-37)
src/plugin/transform/model-resolver.test.ts (2)
src/plugin/transform/model-resolver.ts (1)
resolveModelWithTier(163-270)src/plugin/transform/index.ts (1)
resolveModelWithTier(22-22)
src/plugin/transform/gemini.test.ts (2)
src/plugin/transform/index.ts (1)
isImageGenerationModel(52-52)src/plugin/transform/gemini.ts (1)
isImageGenerationModel(152-158)
🔇 Additional comments (7)
src/plugin/transform/model-resolver.ts (1)
55-57: LGTM — alias mappings follow the existing Opus pattern exactly.All three tier variants (
low,medium,high) correctly resolve to"claude-sonnet-4-6-thinking", and the tier suffix feeds intoTHINKING_TIER_BUDGETS.claude(8192 / 16384 / 32768) through the existing resolution path.src/plugin/config/models.ts (1)
75-83: LGTM — model definition mirrors the Opus entry structure.Context/output limits, modalities, and variant budgets (
low: 8192,max: 32768) are consistent with both the Antigravity API spec and the Opus model. ThemaxOutputTokensceiling (64000) stays above the max thinking budget (32768), satisfying the API constraint.README.md (1)
198-206: LGTM — config example correctly lists "low" and "max".src/plugin/transform/cross-model-sanitizer.test.ts (1)
16-16: LGTM — assertion is consistent with theisClaudeModelimplementation.src/plugin/transform/claude.test.ts (1)
63-63: LGTM — mirrors the existing 4.5 assertion and correctly validates the new model name.docs/MODEL-VARIANTS.md (1)
111-119: LGTM — example JSON is correct and matches themodels.tsdefinition.src/plugin/config/models.test.ts (1)
20-20: LGTM — alphabetical placement is correct.
"antigravity-claude-sonnet-4-6-thinking"sits correctly between"antigravity-claude-sonnet-4-6"(shorter string sorts first) and"antigravity-gemini-3-flash"(c<g).
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@docs/ANTIGRAVITY_API_SPEC.md`:
- Line 82: The table row for the new model `claude-sonnet-4-6-thinking` was
added but the Changelog was not updated; add a new Changelog entry dated today
under the "Changelog" section noting the addition of the
`claude-sonnet-4-6-thinking` model (include version or release note text
consistent with existing entries), mirror the format used by existing changelog
lines, and ensure the entry references the new model ID exactly so the docs
remain consistent.
In `@README.md`:
- Line 124: Update the README table entry for the model named
"antigravity-claude-sonnet-4-6-thinking" so its Variant column matches the
actual configured variants (`low` and `max`) defined in models.ts (see the
variant definitions around the symbols that map the model's thinking budgets at
lines referenced as variants in models.ts). Replace "low, medium, high" with
"low, max" in that table row so CLI users who pass --variant get the correct
documented options.
In `@src/plugin/transform/model-resolver.test.ts`:
- Around line 130-136: The test and resolver are still asserting the deprecated
`thinkingBudget` (budget_tokens) behavior for Sonnet/Opus 4.6; update the
resolver `resolveModelWithTier` and the corresponding test
(`antigravity-claude-sonnet-4-6-thinking` case in model-resolver.test.ts) to
return the new adaptive thinking representation and output effort instead of a
numeric thinkingBudget — i.e., produce something like a `thinking` object with
`type: "adaptive"` and an `output_config.effort` value (e.g., "high") and update
assertions to check for those fields (`result.thinking.type`,
`result.output_config.effort`, and keep
actualModel/isThinkingModel/quotaPreference checks), and open a follow-up issue
to remove `thinkingBudget` usages across the codebase once Anthropic fully
deprecates budget_tokens.
---
Nitpick comments:
In `@docs/MODEL-VARIANTS.md`:
- Around line 163-172: Update the "Tier-suffixed model names" list to include
the Sonnet 4.6 Thinking variants that the resolver supports: add entries for
antigravity-claude-sonnet-4-6-thinking-low,
antigravity-claude-sonnet-4-6-thinking-medium, and
antigravity-claude-sonnet-4-6-thinking-high (matching the same path as the Opus
variants) so users relying on tier-suffixed strings see they are accepted.
In `@src/plugin/config/models.test.ts`:
- Around line 52-57: Add an assertion for the new Sonnet thinking model so its
variants are directly validated: update the "defines thinking budget variants
for Claude thinking models" test (which calls getModel) to also expect
getModel("antigravity-claude-sonnet-4-6-thinking").variants toEqual the same
shape as the Opus one ({ low: { thinkingConfig: { thinkingBudget: 8192 } }, max:
{ thinkingConfig: { thinkingBudget: 32768 } } }); this ensures both
antigravity-claude-opus-4-6-thinking and antigravity-claude-sonnet-4-6-thinking
are explicitly checked.
In `@src/plugin/transform/gemini.test.ts`:
- Around line 615-618: Test title says "thinking variants" (plural) but only
asserts one variant; update the spec in src/plugin/transform/gemini.test.ts for
the isImageGenerationModel checks: either change the it(...) title to singular
("thinking variant") or add an additional assertion such as
expect(isImageGenerationModel("claude-sonnet-4-6-thinking-low")).toBe(false) so
the title and assertions match; reference the isImageGenerationModel calls in
the failing test block and adjust accordingly.
In `@src/plugin/transform/model-resolver.test.ts`:
- Around line 121-137: Add test cases to cover the new Sonnet 4.6 thinking
aliases: ensure resolveModelWithTier("antigravity-claude-sonnet-4-6-thinking")
and the tier-suffixed aliases
("gemini-claude-sonnet-4-6-thinking-low/medium/high") map to actualModel
"claude-sonnet-4-6-thinking", have thinkingBudget 32768, isThinkingModel true,
and quotaPreference "antigravity". Mirror the existing Opus tests: add the
non-thinking Sonnet test into the "Claude Sonnet 4.6 (non-thinking)" describe
block and add the tier-suffixed backward-compatibility tests into the "backward
compatibility" describe block so resolveModelWithTier and the new aliases in
model-resolver.ts are exercised.
| | Model Name | Model ID | Type | Status | | ||
| |------------|----------|------|--------| | ||
| | Claude Sonnet 4.6 | `claude-sonnet-4-6` | Anthropic | ✅ Verified | | ||
| | Claude Sonnet 4.6 Thinking | `claude-sonnet-4-6-thinking` | Anthropic | ⏳ Pending Rollout | |
There was a problem hiding this comment.
LGTM — model row is consistent with existing table format.
The new entry correctly uses the API model ID claude-sonnet-4-6-thinking, matching the convention for other Claude rows.
Note: the Changelog section (lines 634-635) was not updated to record this addition. Consider adding an entry dated today.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@docs/ANTIGRAVITY_API_SPEC.md` at line 82, The table row for the new model
`claude-sonnet-4-6-thinking` was added but the Changelog was not updated; add a
new Changelog entry dated today under the "Changelog" section noting the
addition of the `claude-sonnet-4-6-thinking` model (include version or release
note text consistent with existing entries), mirror the format used by existing
changelog lines, and ensure the entry references the new model ID exactly so the
docs remain consistent.
| | `antigravity-gemini-3.1-pro` | low, high | Gemini 3.1 Pro with thinking (rollout-dependent) | | ||
| | `antigravity-gemini-3-flash` | minimal, low, medium, high | Gemini 3 Flash with thinking | | ||
| | `antigravity-claude-sonnet-4-6` | — | Claude Sonnet 4.6 | | ||
| | `antigravity-claude-sonnet-4-6-thinking` | low, medium, high | Claude Sonnet 4.6 with extended thinking | |
There was a problem hiding this comment.
Variant list is wrong — "low, medium, high" should be "low, max".
The model table documents variants that don't exist in the definition. The actual variants (lines 202–205 below and models.ts lines 80–81) are low and max. Users reading the table who try --variant=medium or --variant=high will receive no configured thinking budget, silently falling back to the model's default (32768). The Opus row on the very next line correctly shows "low, max".
📝 Proposed fix
-| `antigravity-claude-sonnet-4-6-thinking` | low, medium, high | Claude Sonnet 4.6 with extended thinking |
+| `antigravity-claude-sonnet-4-6-thinking` | low, max | Claude Sonnet 4.6 with extended thinking |📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| | `antigravity-claude-sonnet-4-6-thinking` | low, medium, high | Claude Sonnet 4.6 with extended thinking | | |
| | `antigravity-claude-sonnet-4-6-thinking` | low, max | Claude Sonnet 4.6 with extended thinking | |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@README.md` at line 124, Update the README table entry for the model named
"antigravity-claude-sonnet-4-6-thinking" so its Variant column matches the
actual configured variants (`low` and `max`) defined in models.ts (see the
variant definitions around the symbols that map the model's thinking budgets at
lines referenced as variants in models.ts). Replace "low, medium, high" with
"low, max" in that table row so CLI users who pass --variant get the correct
documented options.
| it("antigravity-claude-sonnet-4-6-thinking gets default max budget (32768)", () => { | ||
| const result = resolveModelWithTier("antigravity-claude-sonnet-4-6-thinking"); | ||
| expect(result.actualModel).toBe("claude-sonnet-4-6-thinking"); | ||
| expect(result.thinkingBudget).toBe(32768); | ||
| expect(result.isThinkingModel).toBe(true); | ||
| expect(result.quotaPreference).toBe("antigravity"); | ||
| }); |
There was a problem hiding this comment.
🧩 Analysis chain
🌐 Web query:
Anthropic Claude Sonnet 4.6 budget_tokens deprecated effort parameter thinking
💡 Result:
On Claude Sonnet 4.6 (claude-sonnet-4-6):
thinking: { type: "enabled", budget_tokens: N }is deprecated on Sonnet 4.6 (and Opus 4.6) and “will be removed in a future model release.” The recommended replacement is adaptive thinking +effort. [1]- Use adaptive thinking by setting:
thinking: { type: "adaptive" }. [1] - Control how much thinking / overall token spend Claude uses with the
effortparameter viaoutput_config, e.g.low | medium | high(default). (maxis Opus 4.6 only; usingmaxon Sonnet 4.6 errors.) [1][2] effortis not deprecated; it’s the recommended control on Sonnet 4.6/Opus 4.6 in place ofbudget_tokens. [2]
Example (Sonnet 4.6, recommended):
{
"model": "claude-sonnet-4-6",
"max_tokens": 16000,
"thinking": { "type": "adaptive" },
"output_config": { "effort": "medium" },
"messages": [{ "role": "user", "content": "..." }]
}[1][2]
Claude Sonnet 4.6 uses a deprecated thinking budget approach — track for migration to effort parameter.
Per Anthropic's official documentation, budget_tokens on Sonnet 4.6 (and Opus 4.6) is deprecated and will be removed in a future model release. The recommended replacement is adaptive thinking with the effort parameter (low | medium | high).
The test correctly validates the current code path, but the thinkingBudget approach it validates is deprecated. Before Anthropic removes the budget_tokens API, migrate to:
{ "thinking": { "type": "adaptive" }, "output_config": { "effort": "high" } }Consider opening a follow-up issue to update the resolver and test to use the effort parameter approach instead.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@src/plugin/transform/model-resolver.test.ts` around lines 130 - 136, The test
and resolver are still asserting the deprecated `thinkingBudget` (budget_tokens)
behavior for Sonnet/Opus 4.6; update the resolver `resolveModelWithTier` and the
corresponding test (`antigravity-claude-sonnet-4-6-thinking` case in
model-resolver.test.ts) to return the new adaptive thinking representation and
output effort instead of a numeric thinkingBudget — i.e., produce something like
a `thinking` object with `type: "adaptive"` and an `output_config.effort` value
(e.g., "high") and update assertions to check for those fields
(`result.thinking.type`, `result.output_config.effort`, and keep
actualModel/isThinkingModel/quotaPreference checks), and open a follow-up issue
to remove `thinkingBudget` usages across the codebase once Anthropic fully
deprecates budget_tokens.
|
hey @sk0x0y do you know why only low and max supported by google ? like we don't even see the option of that on anti gravity |
…emini-3-pro - Add medium/high/max thinking variants for Claude Opus 4.6 (low/medium/high/max) - Add low/medium/high thinking variants for Claude Sonnet 4.6 Thinking - Add 'max' to ThinkingTier type, TIER_REGEX, and THINKING_TIER_BUDGETS - Remove antigravity-gemini-3-pro model definition - Remove gemini-3-pro-preview CLI model definition - Update all aliases, tests, docs, and scripts - Include claude-sonnet-4-6-thinking model support (supersedes PR #496) Supersedes: #496
|
superseeded in #503, thanks for the groundwork! |
…emini-3-pro - Add medium/high/max thinking variants for Claude Opus 4.6 (low/medium/high/max) - Add low/medium/high thinking variants for Claude Sonnet 4.6 Thinking - Add 'max' to ThinkingTier type, TIER_REGEX, and THINKING_TIER_BUDGETS - Remove antigravity-gemini-3-pro model definition - Remove gemini-3-pro-preview CLI model definition - Update all aliases, tests, docs, and scripts - Include claude-sonnet-4-6-thinking model support (supersedes PR #496) Supersedes: #496
3 participants
Summary
Adds support for
antigravity-claude-sonnet-4-6-thinkingmodel with extended thinking capabilities.Changes
antigravity-claude-sonnet-4-6-thinkingmodel definition withthinkingConfigvariants (matching existing Opus 4.6 Thinking pattern):low:thinkingBudget: 8192max:thinkingBudget: 32768Model Definition
Variant Options
Following the existing
antigravity-claude-opus-4-6-thinkingpattern:low:thinkingBudget: 8192max:thinkingBudget: 32768Evidence
Google Antigravity now supports Claude Sonnet 4.6 (thinking) as an official reasoning model:
Use Case
Enables users to use thinking capabilities with Sonnet 4.6:
Related
Testing